CAJ | 학술논문

超分辨率（Super Resolution，SR）是一种广泛用于提高图像分辨率的技术，现实生活的许多应用都需要单一图像的超分辨率。大多传统超分辨率技术都是从大量外部数据库中抽取训练样例集来训练一个过完备字典对图像块进行稀疏表示。对传统单一图像超分辨率技术进行改进从而提出一种新的方法，它没有用到外部数据库。首先从低分辨率输入图像的双立方插值图像中抽取图像块并作为字典学习的自身样例，从而大大提高训练效率，然后使用自适应稀疏域选择策略（ ASDS）为每个图像块自适应选择子字典，最后在重建过程中加入非局部自相似性约束作为正则化项。大量的实验结果证明它以一种较好的方式提高效率的同时改善了视觉效果。
초분변솔（Super Resolution，SR）시일충엄범용우제고도상분변솔적기술，현실생활적허다응용도수요단일도상적초분변솔。대다전통초분변솔기술도시종대량외부수거고중추취훈련양례집래훈련일개과완비자전대도상괴진행희소표시。대전통단일도상초분변솔기술진행개진종이제출일충신적방법，타몰유용도외부수거고。수선종저분변솔수입도상적쌍립방삽치도상중추취도상괴병작위자전학습적자신양례，종이대대제고훈련효솔，연후사용자괄응희소역선택책략（ ASDS）위매개도상괴자괄응선택자자전，최후재중건과정중가입비국부자상사성약속작위정칙화항。대량적실험결과증명타이일충교호적방식제고효솔적동시개선료시각효과。
The super - resolution( Super Resolution,SR) is a widely used technique to improve the resolution of an image. Many applications in real - life need of single image super - resolution. Most of the conventional super - resolution technology colleot training examples from the large number of outside database for training a over - complete dictionary,which can sparse representation of the image block. In this paper,we improve the traditional single image super - resolution techniques to propose a new algorithm,which does not use an external database. First extracting image blocks from the bicubic interpolation version of the input low -resolution image as the self - sample for learning a dictionary,so that we can greatly improve the efficiency of the training process. An adaptive sparse domain selection strategy(ASDS)for each image block to select a sub -dictionary adaptively is used,The non - local structure self - similarity constraint as a regularization term in the reconstruction process is combined. A large number of experimental results show it in a better way to increase efficiency with improving the visual effect.